Real-Time Interfacial Pressure Prediction in CMP Using Machine Learning Surrogates of Finite Element Simulations

Achieving uniform material removal in chemical-mechanical polishing requires precise tuning of interfacial pressure. High-fidelity finite-element simulations can accurately model pressure profiles essential for this tuning process. However, these simulations are too computationally intensive for extensive parameter studies. This work introduces a surrogate modelling approach for two-dimensional radial pressure profiles in five-zone polishing heads. Functional principal component analysis identifies a compact, data-driven basis from a dataset of finite-element-generated pressure profiles. Kernel ridge regression then maps applied zone loadings to the extracted coefficients. Validation across diverse loading scenarios shows that the proposed surrogate matches the accuracy of detailed finite-element simulations while delivering millisecond-level predictions and clear interpretability which inputs drive specific pressure‐profile changes. Researchers with existing high-fidelity models can adopt this framework to develop fast, interpretable surrogates for real-time process optimization.